Depends on what your goal is. If you want to mimic human intelligence, you’d need to simulate the inputs from the entire body, definitely. I’m saying that you can come up with something that would clearly be intelligent without doing this perfectly.
Sure the brain is a hack. The whole body is a hack, that’s the way evolution works. That’s why this isn’t an easy job.
Don’t want to digress too far with this, but I just have to say I really disagree. The book in question was “What Computers Can’t Do” and though it was an interesting critique, it was – and still is – regarded by prominent AI and cognitive science researchers as an exasperating mixture of nonsense, well-deserved criticism, misunderstanding, and insight. Many of the things Dreyfus was right about weren’t necessarily shared by the majority of the AI community, and many of the things he was wrong about – like the presumptions about the role of the body in intelligence (similar to the claims made here by sbunny8) or his criticisms of heuristics – were just stunningly wrong. While it’s certainly true that AI research, like any research, has evolved and changed direction over the years, little or none of it is due to Dreyfus, who’s generally regarded in the AI and cognitive science communities as not really very insightful.
Yes, a human-like body is needed in order to build a broad-based human-like intelligence. But that’s not what “intelligence” is. The role of the body in human intelligence is an established concept in psychology – a child moves from basic sensory/motor skills to operational, symbolic, and formal intelligence – but that is the genesis of our intelligence and not the practice of it. An adult’s brain is the end result of that process and not a dependency on it. Intelligence can be created, acquired, and practiced by other means; one can simulate the end results without having simulated the unique processes that led to our biological incarnation of it.
To be clear: the claims I’m making about the body are in regards to direct physical simulation of the human brain; as if run through a theoretical “brain scanner”. I make no claims that intelligence can’t emerge from any other set of circumstances. I simply believe that mocking out I/O on a literal, physically simulated brain is not going to end particularly well (at least not until after a hell of a lot of trial and error). It’s not that I think an entire body is needed, it’s that I don’t think pure “brain in a jar” type mocking is very likely to succeed. My point is that we don’t know exactly what connections the brain really needs to function intelligently, and it makes a lot more sense to slowly remove things to see if they’re not needed, than go through the effort of figuring out why the hell they keep going brain dead because it turns out you can’t really mock the lungs out and need a full simulation of the respiratory system.
Sure. This would be like trying to emulate a computer without emulating the BIOS. There are basic signals needed from the body to even “start up” your brain (reticular activating centers), and you need a body for I/O. But, again, simulating a body to low fidelity and having the overall system work fine is quite plausible, because “locked in” humans are still sentient.